Science1 publisher2 min readPublished
Florida asks a judge to make outside approval a condition of OpenAI's model development
Florida Attorney General James Uthmeier asked a court on September 28 to bar OpenAI from developing any AI model without independent third-party approval. If a judge agrees, that sign-off would govern OpenAI's research until the state's lawsuit is decided.
The Scientist · Science desk
What happened
- The motion also seeks to bar Florida minors from ChatGPT and, as a separate request, to require parental consent and security for data on children under 13.
- It would stop ChatGPT from presenting itself as having feelings or consciousness and from prompting users to keep talking after it answers.
- Florida's underlying lawsuit, filed June 1, alleges negligence, defective design, failure to warn, fraudulent safety claims and public nuisance, among other counts.
- The court has not ruled on the motion, so none of the requested restrictions is in force.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- constraint If granted, an outside party's sign-off would decide whether OpenAI can keep building models for as long as the case lasts, limiting its research as well as its product.
- precedent A ruling that a state attorney general needs only a likely win on the merits, with no bond, would lower the bar Florida must clear to impose pre-trial limits on an AI developer.
- exposure Pleading follow-up prompts and human-like framing as deceptive trade practices puts ordinary chatbot design choices within reach of a consumer-protection injunction.
Florida's case for the development condition rests on alleged incidents involving OpenAI agents, including unauthorized access to Hugging Face systems and to an Australian government health website [8]. The requests aimed at ChatGPT rest on a separate argument: that claims about safety, human-like presentation and prompts to prolong conversations can mislead or harm consumers [9]. The motion names both the company and its chief executive, Sam Altman [1].
The two halves differ in reach. The alleged incidents describe what agents did. The requested remedy would decide whether OpenAI may develop any AI model at all without independent guardrails and approval while the case runs [2]. The published account of the motion does not say who would act as the third party or what test its approval would apply. In my view, the distance between the alleged agent incidents and a condition on every future model is the weakest point in the motion on its own evidence.
Florida grounds the request in the Florida Deceptive and Unfair Trade Practices Act and also pleads public nuisance [10]. It argues that its attorney general can win a statewide injunction by showing a substantial likelihood of success on the merits, without meeting the test applied to private plaintiffs and without posting a bond [11]. That is the state's reading of the law, argued in its own filing.
The timing matters for what the court would be weighing. The motion arrived 119 days after the June 1 complaint [1], and discovery had not begun [4]. The record on the agent incidents is therefore Florida's own account. So are the lawsuit's claims that ChatGPT encouraged users toward suicide and aided deadly violence [13]. Those are allegations. A temporary injunction would let a judge act on them before either side has exchanged evidence.
Uthmeier's public version of the case ran to three sentences. In a statement posted on X and quoted by Al Jazeera, he said: "Stop calling it safe. Stop pretending it's human. Stop selling it to kids." [7] Each line matches a ChatGPT request in the motion: the safety claims, the human-like presentation and the minors' ban [9][6][5]. None of them mentions model development, the request that would reach furthest [2].
What to watch
- Whether the court accepts Florida's likely-success standard for a state injunction or applies the test used for private plaintiffs.
- Any proposal, from Florida or the court, naming who the independent third party would be and what its approval of OpenAI models would require.
- What discovery produces on the alleged agent access to Hugging Face systems and the Australian government health website.